Morandi Cinematic Poster · Zeejay
作者 / Author: Zeejay0
Modified edition by Shirehuhu (2026-10-14; shared 2026-10-15). Original copyright, source and attribution: see NOTICE. Governed by the unchanged upstream Personal Non-Commercial License in LICENSE. This is not an official upstream release.
Photography, title and small type form one composition. Typography—not added graphic symbols—creates the cinematic hierarchy. A personal photo may evoke key art without pretending to be an actual production.
Photographic plate
For portrait inputs default to exact 3:4; for landscape/non-portrait retain the source ratio unless overridden. Use the smallest safe crop from low-information edges; protect faces, poses, props, landmark identity, directional axes and useful negative space. Never stretch or cut a principal subject just to fit.
Perform requested mild natural color/white-balance correction before locking the plate, consistent with the user's “先调色” preference. Then treat that prepared plate as LOCKED SOURCE PHOTO. Outside explicit glyph/occlusion masks, retain geometry, perspective, framing, people/vehicles/signage/clouds/windows, exposure, contrast, noise and detail. No further repainting, relighting, retouching, blur, sharpening, denoising, aging or global paper overlay.
A separately requested 原图 comparison panel uses the untouched source, not this graded plate; identify a graded panel as 调色照片.
Use assets/reference-typography-material-v1.png only for restrained ink/paper material cues, never its wording or layout.
Composition and copy
Map the hero, semantic recognition zones, gaze/gesture, first three attention points, quiet type fields, scene axes/vanishing lines, busy exclusions and palette. Infer one short scene-derived premise for mood/copy; invent no new visual events or people.
Choose one theatrical family: bottom-anchor, top-proclamation, central-emblem, integrated-axis, or deliberate edge-scale with recoverable words. Build one dominant cluster, one support cluster and a quiet zone; do not decorate all four edges.
Only the original scene-derived title is mandatory, normally 1–4 words. Select supporting roles only when space and hierarchy benefit:
- identity/byline: supplied name/alias, non-personal study phrase, or obviously fictional creator/character;
- tagline: one emotional proposition, usually 3–10 words in at most two lines;
- atmosphere: source-grounded nouns, known place/weather/time, or an explicitly fictional chapter/archive code;
- compact optional cinematic microcopy: scene nouns, original fictional roles/names/imprint or design-study language;
- action marker: supplied/known date, edition/chapter code or a concise cinematic-study phrase.
Invented credits/codes must be coherent and clearly fictional, not claims about the photo or a real production. No unrequested real identities, studios, awards, ratings, logos, laurels or certification marks. Never copy a named film's title lettering/campaign. Across a series vary copy, line breaks, family and occlusion, not the underlying quality.
Original title and hierarchy
Use at most two type families: title plus support; any mono/condensed microtype belongs to the support system. Choose scene-appropriate serif/Didone, compressed grotesk/slab, tracked geometric sans, humanist/softened face or an original hybrid.
Modify only 1–3 meaningful glyph details derived from a roofline, window rhythm, road, shadow or silhouette. Preserve immediate recognition; avoid distorting every letter, franchise lettering, metal/neon/glow/bevel/grunge gimmicks.
The title must read within three seconds at thumbnail size. Make line breaks semantic/rhythmic, avoid accidental orphans, and align width with a real shoulder/facade/horizon/opening/credit boundary or deliberate bleed. Cropped letters must remain recoverable.
With title cap height = 100, starting ranges are: identity/action 20–32; tagline/context 12–20; atmosphere 8–12; microcopy 5–9. Keep adjacent levels distinct, one dominant L1 and normally no more than one L2 cluster. Use three levels for a sparse teaser, four when useful, five only when the photograph supports density; do not force every role.
Type–image integration
- Protect faces, eyes, mouth, key gestures, landmark crowns, ornaments, readable source signage and product marks. No crossbar/counter/microtype across identifying features.
- Put small type on calm detail with contrast and breathing room. A large title may bridge regions only if critical glyphs remain readable.
- Follow or deliberately counter a real horizon, facade, curb, window rhythm, light beam or perspective axis. Keep baselines optically level unless a diagonal is justified.
- Use 1–3 meaningful hero/title occlusions at most: title on a coherent plane, selected hero edges in front, selected strokes returning over calm areas. No fragmented mask weave; small type stays fully readable.
- Sample muted title inks from the scene or a restrained complement. Legibility comes from local value/position/weight/opacity, not boxes, outlines, shadows or glow. Most support type uses one quiet neutral, with at most one action accent.
- Verify both thumbnail readability and grayscale value separation.
Identity text uses lighter uppercase/small caps and measured tracking; 1–3 names/phrases remain one subordinate line or balanced group. Tagline is a quiet voice, not a second headline. Optional billing-like microcopy uses legible condensed support type, tight controlled leading, clean block edges and alignment to title/axis; role/name contrast only when the renderer can reproduce it. Keep it off faces, cars, paving/foliage or other busy detail. Separate any action marker with breathing room.
Material and execution
Use a 2–3-color low-saturation type palette. Broad fibers/shallow letterpress may occur inside large glyphs; small glyphs use clean ink/finer grain. Every material effect ends at the glyph edge. No global texture, panels, rectangles, circles, frames, stickers, tape, decorative stars/symbols or unrelated graphics.
Layer locally: locked photo → title/support type → selected hero edges → selected returning title strokes → readable small type. Use the built-in image editing capability with the actual source; consult imagegen for tool mechanics. Do not send private references to unrelated services.
The brief must specify plate preparation/crop/invariants; hero protection and scene axes; chosen family; exact title and every included support line; font/category/customization; hierarchy/line breaks/positions/ink/contrast; planned occlusions; glyph-only texture and relevant exclusions. Do not require unused text roles or decorative elements merely to fill a template.
Inspect the actual result against the locked plate, complete subject, exact text, one clear hierarchy/family, two-family limit, clean occlusions, thumbnail/grayscale readability and glyph-only texture. Correct localized failures; after two failed correction attempts explain the limitation rather than claim it passed.
Export at least 2K long edge, proportionally; disclose enlargement, never invented native detail. Preserve originals/prior outputs and basenames in separate result folders. Original/result comparison remains a separately confirmed output. Return the image/path and a short explanation; detailed prompt only when requested. No automatic promotion or generation-count notices.